Jensen Huang Confirms US Rules Limit Blackwell Chips in China

TLDRs:

• Nvidia CEO Jensen Huang confirms Blackwell chips cannot be sold to China.
• TSMC wafer production remains key to Nvidia’s AI chip expansion.
• SK Hynix, Samsung, and Micron support Nvidia with advanced memory chips.
• Packaging and memory constraints may keep supply tight through 2025.

Nvidia CEO Jensen Huang has reiterated that US export restrictions prevent the company from selling its next-generation Blackwell AI chips to China, even as global demand for the technology continues to surge.

Huang made the remarks during an event in Hsinchu, Taiwan, highlighting both the opportunities and constraints facing Nvidia’s ambitious AI hardware plans.

Strong Global Demand for Blackwell Chips

According to Huang, Nvidia is seeing “strong demand” for its Blackwell chips, which are poised to power a new generation of AI workloads and high-performance computing tasks.

While he did not provide specific order numbers, the company’s reliance on advanced chip manufacturing and memory technology suggests that production will be pushed to its limits in the near term.

The Blackwell chips represent a significant step up from Nvidia’s prior H100 GPUs, both in performance and power requirements, with the GB200 NVL72 rack-scale systems drawing up to 120 kilowatts, roughly ten times the power of earlier CPU racks.

TSMC Remains Crucial Supplier for Nvidia

Huang emphasized that Nvidia’s success is closely tied to Taiwan Semiconductor Manufacturing Company (TSMC), the firm producing the wafers necessary for the Blackwell chips.

This marks Huang’s fourth trip to Taiwan in 2025, underscoring the critical partnership. TSMC CEO C.C. Wei confirmed that Nvidia has requested additional wafers, but details of the production schedule remain confidential.

The company’s advanced CoWoS (Chip-on-Wafer-on-Substrate) technology, which integrates multiple compute and memory chips on a single substrate, is expected to reach 90,000–95,000 wafers per month by the end of 2026, up from 70,000 at the end of 2025. Even with this increase, supply bottlenecks are anticipated to continue through the remainder of 2025.

Memory Makers Boost Production Capacity

In addition to wafer constraints, memory chip availability is another critical factor. Huang noted that SK Hynix, Samsung, and Micron have provided advanced memory samples and expanded production capacity to support Nvidia’s Blackwell deployments.

SK Hynix reported that its 2026 chip output is already sold out and has committed to additional investments to increase capacity, while Samsung is negotiating to supply next-generation memory modules. The combination of CoWoS packaging and High Bandwidth Memory (HBM) creates a bottleneck that could influence the pace of global AI infrastructure rollout.

US Export Rules Block China Shipments

Perhaps most notably, Huang confirmed that there are no ongoing discussions to sell Blackwell chips to China due to US export regulations.

The restrictions form part of broader technology controls aimed at limiting the transfer of advanced semiconductor capabilities. This policy effectively ensures that Nvidia’s most cutting-edge AI hardware will remain outside the Chinese market for the foreseeable future, influencing both global competition in AI computing and regional supply chains.

Despite these limitations, Nvidia remains focused on scaling its Blackwell platform through partnerships with colocation providers, hyperscale operators, and enterprise data centers worldwide. Innovations such as the modular MGX architecture allow operators to scale upgrades incrementally, helping smaller data centers remain competitive with large cloud providers.

Still, with packaging constraints, wafer production limits, and high memory demand, analysts anticipate supply tightness to persist through 2025, with gradual relief expected in 2026 as production ramps up.

The post Jensen Huang Confirms US Rules Limit Blackwell Chips in China appeared first on CoinCentral.